Anomalous Sound Detection using Multi -Feature Representation and Domainmixed GMM

In the real-world manufacturing environment, machine anomalies occur infrequently, and these anomalies are often complex and diverse, making them difficult to detect. Various factors, such as different operational phases, varying machine conditions, and changes in the surrounding environment, can lead to unpredictable variations in machine sound characteristics. As a result, normal sounds can easily be mistaken for abnormal ones. Therefore, there is a need to develop an ASD (Anomaly Sound Detection) system with strong domain generalization capabilities. By enhancing domain generalization ability, the model can demonstrate better adaptability, regardless of how machine sound characteristics may change. In this study, within a single neural network, multiple feature representation methods are employed to learn and extract features from the data, capturing the characteristics of different domains. Multiple subsystems are utilized to process different feature representations, followed by data augmentation to combine data from various domains, ultimately enhancing the system's generalization capabilities.